Abstract. Creators are widely advised to optimize when they post. We test whether posting time predicts short-form video performance using a 23,256,969-video index of public TikTok content, of which 12,683,551 videos by 189,718 creators (all creators with at least eight indexed videos) enter the analysis. Using a within-creator design that controls for audience size, follower loyalty, and niche, we find that posting hour and day of week jointly explain 0.007% of the variance in a video's performance relative to the creator's own baseline (R² = 0.000069). The gap between the best and worst posting hour of the day is 0.86 percentile points (creator-clustered 95% CI [0.62, 1.11]), and a snapshot-age diagnostic suggests even this is an upper bound. Personalized "best hours" fail an out-of-sample split-history test, posting-volume swings of 2.7× across hours leave outcomes unmoved, and machine-scheduled posts slightly outperform organically timed ones. In a same-sample ablation on a separate 39,414-post corpus, adding machine-extracted content attributes to a timing-and-account baseline raises AUC from 0.532 to 0.574 under identical account-grouped folds (0.624 against a views-based label). Relative to timing, content attributes carry several times the recoverable signal - though most within-creator variance is explained by neither, consistent with popularity being dominated by factors beyond observable scheduling and surface features.
The abstract above states the result. Here is what the rest of the paper covers:
Roughly 5,000 words, with every table and figure.
Free. We will email you the working paper, the full report and the deck. Layers will not sell, lease or transfer your email to any other company.